Improve uplift_at_k validation and add top-k markers to prediction plots

[RiukaRin]

This PR improves metric robustness and visualization usability in three related areas:

Added explicit validation in uplift_at_k(..., strategy="overall") to ensure the selected top-k slice contains both treatment and control samples.
Previously, this edge case could return nan
Now it raises a clear ValueError with guidance
Extended plot_uplift_preds with a new optional k parameter
Supports k as float in (0, 1) or int in [1, n_samples].
Draws top-k threshold markers (vertical lines) for treatment, control, and uplift histograms.
Adds validation for invalid k values/types.
Tightened metric tests by replacing broad pytest.raises(Exception) with explicit pytest.raises(ValueError) where appropriate.
Also updated API docs for plot_uplift_preds to mention top-k threshold markers.

Verification Process
Validated syntax of updated files via Python compile checks (py_compile)
Ran direct runtime checks for new uplift_at_k behavior:

• Confirmed ValueError is raised for top-k slices missing one group in overall strategy
• Confirmed valid scenarios still return finite values.

Added/updated tests:
New test for uplift_at_k empty-group edge case
New/updated tests for plot_uplift_preds(k=...), including invalid k cases
Updated exception-type assertions in metric tests to specific ValueError

Note: full pytest run is pending in a project-compatible environment (not available in the current environment due to pytest/scikit-learn setup mismatch for full test execution)

Additional info
This PR is intentionally structured as three logical commits:

1. metric validation (uplift_at_k)
2. visualization feature (plot_uplift_preds + docs/tests)
3. stricter exception-type tests in metrics